Machine Learning-Based Models for Early Stage Detection of Autism Spectrum Disorders

Abstract

Autism Spectrum Disorder (ASD) is a group of neurodevelopmental disabilities that are not curable but may be ameliorated by early interventions. We gathered early-detected ASD datasets relating to toddlers, children, adolescents and adults, and applied several feature transformation methods, including log, Z-score and sine functions to these datasets. Various classification techniques were then implemented with these transformed ASD datasets and assessed for their performance. We found SVM showed the best performance for the toddler dataset, while Adaboost gave the best results for the children dataset, Glmboost for the adolescent and Adaboost for the adult datasets. The feature transformations resulting in the best classifications was sine function for toddler and Z-score for children and adolescent datasets. After these analyses, several feature selection techniques were used with these Z-score-transformed datasets to identify the significant ASD risk factors for the toddler, child, adolescent and adult subjects. The results of these analytical approaches indicate that, when appropriately optimised, machine learning methods can provide good predictions of ASD status. This suggests that it may possible to apply these models for the detection of ASD in its early stages.

Use of Machine Learning to Investigate The Quantitative Checklist For Autism in Toddlers (Q-CHAT) Towards Early Autism Screening

“…To further explore the cross-cultural validity of the instrument we applied the three ML classi ers to the 10 items selected by Allison and colleagues on the Q-CHAT-10 (2012) [19] and we found that, in our sample, SVM algorithm was able to classify autism with 87% accuracy, 65% sensitivity and 86% speci city. These results are in line with those recently reported by Akter et al (2019) [29], where a dataset of Q-CHAT-10 administered using a mobile application [27] was analyzed using ML and the SVM algorithm was able to classify autism with 98% of accuracy. Together, these ndings con rm a satisfactory cross-cultural validity of the Q-CHAT in different samples, countries and languages.…”

“…In the rst study, Thabath and colleagues (2019) [28], analyzed the adult version of the AQ-10 using a new rules-machine learning (RML) and were able to achieve an accuracy of about 90%, sensitivity of 87% and speci city of about 90%, while in the second study, the same authors applied the Naïve Bayes algorithm to the AQ-10 and found a similar accuracy of 92.8%, 91.3% and 95.7% for the child, adolescent and adult versions respectively. To the best of our knowledge, the only study applying ML to toddlers was conducted by Akter and colleagues (2019) [29], who analyzed the Q-CHAT-10 collected using the dataset from the ASDTests app and found that using a range of different classi ers, when optimized, they were able to effectively classify autism with an accuracy of 98%.…”

Identifying predictive features of autism spectrum disorders in a clinical sample of adolescents and adults using machine learning

“…Previous studies have applied machine learning techniques to examine whether the process of diagnosing ASD can be improved by statistically identifying reduced subsets of features from existing diagnostic instruments reaching from self-administered screening questionnaires to clinician-administered diagnostic tools (for a recent overview, see Thabtah 32 ). A few authors have shown that efficiency and accessibility of existing pre-diagnostic screening questionnaires such as the Autism-Spectrum Quotient (AQ) [33][34][35] or the Social Responsiveness Scale (SRS) 36,37 can be improved using machine learning. Similar machine learning experiments have been run to identify subsets of behavioral features from clinician-administered diagnostic tools, namely ADOS (Module 1 to 3) [38][39][40][41][42] and ADI-R 36,39,43 .…”

Classification and Biomarker Exploration of Autism Spectrum Disorders Based on Recurrent Attention Model

“…In the voxel-based algorithms, Ecker et al performed ASD classification according to the local differences of voxel value in white matter and obtained 81% accuracy [17]. Uddin et al divided the original 3D MRI data into several small blocks, and analyzed the features of these small blocks.…”